Solar Pro 4
- Source
- upstage.ai
- Author
- Upstage
- Date
Context
Upstage says Solar Pro 4 was trained around completed office assignments. Its release post states that OfficeVerse, its training pipeline, synthesizes office tasks from real public data across 11 industry domains and 12 task types and grades each one pass or fail on the final deliverable. The design makes the finished work product, rather than a single answer, the thing being graded.
To illustrate the intended workflow, Upstage describes a fictional coffee-store site analysis: the model receives one store-opening policy document and six market-data files and, in three prompts, screens ten candidate sites and produces three deliverables in sequence. Upstage says it is releasing an cookbook with system prompts, actual outputs and pass criteria, including an English version of the assignment readers can run themselves. The training claim and the example are the company's own account; the post's scores come from Artificial Analysis and in-house tests and are not evaluated here.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Checking sign-in…
Loading comments…